1. [para. 1] Ulanqab's data center server rooms, requiring colossal amounts of electricity, provide a climate-controlled sanctuary of constant temperature and humidity, a stark contrast to the intense heat outside.
2. [para. 2] According to the local statistics bureau, the city's electricity consumption from internet data services surged 85.2% year-on-year to a massive 4 billion kilowatt-hours in the first half of 2026.
3. [para. 3] Once the "potato capital," this Inner Mongolian city is rapidly reinventing itself as the "capital of tokens" for AI. By the end of 2025, it had attracted 84 data center projects, mostly intelligent computing centers, with a total investment exceeding 500 billion yuan ($74 billion).
4. [para. 4] Tech giants including Alibaba, Huawei, Apple, and DeepSeek have built computing infrastructure there. Local authorities issued a three-year action plan in 2025 to leverage this into a green-computing foundation for AI model training and inference.
5. [para. 5] Wang Chaoyang of Alibaba Cloud highlighted the abundant cheap green power as a premier advantage. In Ulanqab, green power accounts for 90% of their data center's energy consumption, outpacing their national average of 73.6%.
6. [para. 6][para. 7] Now branding itself as "fertile land for tokens," the city hosts Alibaba Cloud, which opened its first project there in 2020. Demand skyrocketed in late 2023, running their facilities out of capacity. Today, Alibaba Cloud operates nearly half of the city's total data centers.
7. [para. 8][para. 9] Ulanqab became China's largest intelligent computing base, surpassing Zhangjiakou, primarily due to lower power prices (0.3–0.35 yuan/kWh), which can save a 1-gigawatt data center 5 billion yuan annually in electricity costs. Its proximity to Beijing (350 km) provides a network latency of just 4 milliseconds, while its dry climate, average temperature of 4.3°C, and altitude of 1,500 meters offer natural cooling advantages crucial for massive server farms.
8. [para. 10][para. 11] Wang noted that supply-chain bottlenecks migrated upstream from CPUs to storage to power supply, making Ulanqab an ideal option. He observes a geographic bifurcation in China's AI computing ecosystem: AI training moves westward for cheap energy, while AI inference moves eastward for proximity to end users.
9. [para. 12][para. 13] A 1-gigawatt data center is a massive power guzzler, consuming 8.8 billion kWh annually, equivalent to a small city. By 2030, electricity demand from China's data centers could reach 700 billion kWh (5.3% of the national total). Wang noted that while early small chip clusters stayed near the coast, clusters of tens or hundreds of thousands of chips have relocated west of the Hu Huanyong Line to hubs like Ulanqab.
10. [para. 14][para. 15] The speed of infrastructure deployment determines computing power availability. Traditional construction cycles can stretch to 30 months in Europe and 12-24 in the US, but only 6-12 in China. A modular AI data center in Ulanqab can be fully delivered in just 100 days, which significantly accelerates the construction pace while slashing overall costs by more than 10%.
11. [para. 16][para. 17] Wang argued that modular data centers can leverage China's manufacturing capabilities for export. This productized engineering approach mitigates shortages of specialized onsite labor, enables rapid GPU deployment to generate immediate revenue and avoid depreciation losses, and streamlines maintenance through standardized designs.
12. [para. 18][para. 19] Omdia analyst Nian Zijun noted that leveraging modularity to boost productivity is a prevailing trend among top-tier cloud providers. This approach drastically improves delivery speed, locks in core parameters like power usage effectiveness during the prefabrication stage, and forces operations teams to participate in the design phase earlier to avoid rework.
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新